TractorGPT (GPT-2)

Fine-tuned GPT-2 for tractor Q&A style text (TractorJunction-style data).

Intended use

• Continue / polish answers in a Question: … / Answer: format.
• Best used with retrieval over a tractor knowledge base.
• Standalone generation may hallucinate prices or specs — prefer grounded retrieval for facts.

Load

from transformers import GPT2LMHeadModel, GPT2Tokenizer

repo = "ashokdevatwal/tractorgpt-gpt2"
tokenizer = GPT2Tokenizer.from_pretrained(repo)
model = GPT2LMHeadModel.from_pretrained(repo)
tokenizer.pad_token = tokenizer.eos_token

Training

• Base: gpt2
• Objective: causal LM on paraphrased tractor Q&A
• See project config/model_config.json and scripts/train_gpt.py